Dog Breed Rank Prediction using Random Forest Regressor

Best-Fit a Dog for me?

https://www.kaggle.com/datasets/pranav941/hows-that-dog-for-me

Preprocessing

#import neccessary libraries
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import plotly.express as px
import seaborn as sns
from plotly.subplots import make_subplots
import plotly.graph_objects as go
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error, r2_score
import math

#import dataset
df_rank = pd.read_csv("rank.csv")
df_trait = pd.read_csv("traits.csv")

df_rank.info()
print()
df_rank.head()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 198 entries, 0 to 197
Data columns (total 9 columns):
 #   Column     Non-Null Count  Dtype  
---  ------     --------------  -----  
 0   Breed      198 non-null    object 
 1   2013 Rank  177 non-null    float64
 2   2014 Rank  184 non-null    float64
 3   2015 Rank  184 non-null    float64
 4   2016 Rank  189 non-null    float64
 5   2017 Rank  190 non-null    float64
 6   2018 Rank  192 non-null    float64
 7   2019 Rank  195 non-null    float64
 8   2020 Rank  195 non-null    float64
dtypes: float64(8), object(1)
memory usage: 14.0+ KB

	Breed	2013 Rank	2014 Rank	2015 Rank	2016 Rank	2017 Rank	2018 Rank	2019 Rank	2020 Rank
0	Retrievers (Labrador)	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
1	German Shepherd Dogs	2.0	2.0	2.0	2.0	2.0	2.0	2.0	3.0
2	Retrievers (Golden)	3.0	3.0	3.0	3.0	3.0	3.0	3.0	4.0
3	Beagles	4.0	5.0	5.0	5.0	6.0	6.0	7.0	7.0
4	Bulldogs	5.0	4.0	4.0	4.0	5.0	5.0	5.0	5.0

df_trait.info()
print()
df_trait.head()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 195 entries, 0 to 194
Data columns (total 17 columns):
 #   Column                      Non-Null Count  Dtype 
---  ------                      --------------  ----- 
 0   Breed                       195 non-null    object
 1   Affectionate With Family    195 non-null    int64 
 2   Good With Young Children    195 non-null    int64 
 3   Good With Other Dogs        195 non-null    int64 
 4   Shedding Level              195 non-null    int64 
 5   Coat Grooming Frequency     195 non-null    int64 
 6   Drooling Level              195 non-null    int64 
 7   Coat Type                   195 non-null    object
 8   Coat Length                 195 non-null    object
 9   Openness To Strangers       195 non-null    int64 
 10  Playfulness Level           195 non-null    int64 
 11  Watchdog/Protective Nature  195 non-null    int64 
 12  Adaptability Level          195 non-null    int64 
 13  Trainability Level          195 non-null    int64 
 14  Energy Level                195 non-null    int64 
 15  Barking Level               195 non-null    int64 
 16  Mental Stimulation Needs    195 non-null    int64 
dtypes: int64(14), object(3)
memory usage: 26.0+ KB

	Breed	Affectionate With Family	Good With Young Children	Good With Other Dogs	Shedding Level	Coat Grooming Frequency	Drooling Level	Coat Type	Coat Length	Openness To Strangers	Playfulness Level	Watchdog/Protective Nature	Adaptability Level	Trainability Level	Energy Level	Barking Level	Mental Stimulation Needs
0	Retrievers (Labrador)	5	5	5	4	2	2	Double	Short	5	5	3	5	5	5	3	4
1	French Bulldogs	5	5	4	3	1	3	Smooth	Short	5	5	3	5	4	3	1	3
2	German Shepherd Dogs	5	5	3	4	2	2	Double	Medium	3	4	5	5	5	5	3	5
3	Retrievers (Golden)	5	5	5	4	2	2	Double	Medium	5	4	3	5	5	3	1	4
4	Bulldogs	4	3	3	3	3	3	Smooth	Short	4	4	3	3	4	3	2	3

#remove \xa0 in column
df_trait['Breed'] = df_trait['Breed'].str.split().str.join(' ')

#combine two dataset based on Breed column
df = pd.merge(df_rank, df_trait, on=['Breed'], how='inner')
df.head()

	Breed	2013 Rank	2014 Rank	2015 Rank	2016 Rank	2017 Rank	2018 Rank	2019 Rank	2020 Rank	Affectionate With Family	...	Coat Type	Coat Length	Openness To Strangers	Playfulness Level	Watchdog/Protective Nature	Adaptability Level	Trainability Level	Energy Level	Barking Level	Mental Stimulation Needs
0	Retrievers (Labrador)	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	5	...	Double	Short	5	5	3	5	5	5	3	4
1	German Shepherd Dogs	2.0	2.0	2.0	2.0	2.0	2.0	2.0	3.0	5	...	Double	Medium	3	4	5	5	5	5	3	5
2	Retrievers (Golden)	3.0	3.0	3.0	3.0	3.0	3.0	3.0	4.0	5	...	Double	Medium	5	4	3	5	5	3	1	4
3	Beagles	4.0	5.0	5.0	5.0	6.0	6.0	7.0	7.0	3	...	Smooth	Short	3	4	2	4	3	4	4	4
4	Bulldogs	5.0	4.0	4.0	4.0	5.0	5.0	5.0	5.0	4	...	Smooth	Short	4	4	3	3	4	3	2	3

5 rows × 25 columns

#Replace NAN with 0
df = df.fillna(0)

Visualization

#create 2 barplots for the first breed "Retrievers (Labrador)"
y = ['Affectionate With Family', 'Good With Young Children', 'Good With Other Dogs', 'Shedding Level', 
     'Coat Grooming Frequency', 'Drooling Level', 'Openness To Strangers', 'Playfulness Level', 
     'Watchdog/Protective Nature', 'Adaptability Level', 'Trainability Level', 'Energy Level', 'Barking Level', 'Mental Stimulation Needs']
fig = px.bar(df[df["Breed"].isin(["Retrievers (Labrador)"])], y=y[0:7], barmode='group')
fig.show()

fig = px.bar(df[df["Breed"].isin(["Retrievers (Labrador)"])], y=y[7:], barmode='group')
fig.show()

#create 2 barplots for the final breed "Pumik"
fig = px.bar(df[df["Breed"].isin(["Pumik"])], y=y[0:7], barmode='group')
fig.show()

fig = px.bar(df[df["Breed"].isin(["Pumik"])], y=y[7:], barmode='group')
fig.show()

#correlation heatmap between variables
sns.heatmap(df.iloc[:,9:].corr(), cmap="PRGn", annot=False, center=0)
plt.show()

df.columns

Index(['Breed', '2013 Rank', '2014 Rank', '2015 Rank', '2016 Rank',
       '2017 Rank', '2018 Rank', '2019 Rank', '2020 Rank',
       'Affectionate With Family', 'Good With Young Children',
       'Good With Other Dogs', 'Shedding Level', 'Coat Grooming Frequency',
       'Drooling Level', 'Coat Type', 'Coat Length', 'Openness To Strangers',
       'Playfulness Level', 'Watchdog/Protective Nature', 'Adaptability Level',
       'Trainability Level', 'Energy Level', 'Barking Level',
       'Mental Stimulation Needs'],
      dtype='object')

#Replace categorical data to numerical
df['Coat Type'].replace(['Double', 'Smooth', 'Silky', 'Curly', 'Wiry', 'Wavy', 'Hairless', 'Rough', 'Corded'],
                        [0, 1, 2, 3, 4, 5, 6, 7, 8], inplace=True)
df['Coat Length'].replace(['Short', 'Medium', 'Long'],
                        [0, 1, 2], inplace=True)

#Histogram visualization
fig = make_subplots(rows=6, cols=3,
                    specs=[[{"colspan": 1}, None, {"colspan": 1}],
                           [{}, {}, {}], 
                           [{}, {}, {}], 
                           [{}, {}, {}],
                           [{}, {}, {}],
                           [{"colspan": 1}, None, {"colspan": 1}]],
                    subplot_titles=("Affectionate With Family", "Good With Young Children", 
                                    "Good With Other Dogs", "Shedding Level", "Coat Grooming Frequency",
                                    "Drooling Level", "Coat Type", "Coat Length",
                                    "Opennes To Strangers", "Playfulness Level", "Watchdog/Protective Nature",
                                    "Adaptability Level", "Trainability Level", "Energy Level",
                                    "Barking Level", "Mental Stimulation Needs",))

fig.add_trace(go.Histogram(x=df["Affectionate With Family"]), row=1, col=1)
fig.add_trace(go.Histogram(x=df["Good With Young Children"]), row=1, col=3)

fig.add_trace(go.Histogram(x=df["Good With Other Dogs"]), row=2, col=1)
fig.add_trace(go.Histogram(x=df["Shedding Level"]), row=2, col=2)
fig.add_trace(go.Histogram(x=df["Coat Grooming Frequency"]), row=2, col=3)

fig.add_trace(go.Histogram(x=df["Drooling Level"]), row=3, col=1)
fig.add_trace(go.Histogram(x=df["Coat Type"]), row=3, col=2)
fig.add_trace(go.Histogram(x=df["Coat Length"]), row=3, col=3)

fig.add_trace(go.Histogram(x=df["Openness To Strangers"]), row=4, col=1)
fig.add_trace(go.Histogram(x=df["Playfulness Level"]), row=4, col=2)
fig.add_trace(go.Histogram(x=df["Watchdog/Protective Nature"]), row=4, col=3)

fig.add_trace(go.Histogram(x=df["Adaptability Level"]), row=5, col=1)
fig.add_trace(go.Histogram(x=df["Trainability Level"]), row=5, col=2)
fig.add_trace(go.Histogram(x=df["Energy Level"]), row=5, col=3)

fig.add_trace(go.Histogram(x=df["Barking Level"]), row=6, col=1)
fig.add_trace(go.Histogram(x=df["Mental Stimulation Needs"]), row=6, col=3)

fig.update_layout(height=1000, showlegend=False)
fig.show()

#Create new variable "Average Ranking" based on average of ranking from 2013 to 2020, then appends it to main dataframe
avg_rank = []
for i in range(0,len(df)):
    temp = 0
    count = 0
    for j in range(1,9):
        if(df.iloc[i,j] != 0):
            temp += df.iloc[i,j]
            count += 1
        else:
            pass
    temp /= count
    avg_rank.append(temp)
    
df['Average Rank'] = avg_rank
df.head()

	Breed	2013 Rank	2014 Rank	2015 Rank	2016 Rank	2017 Rank	2018 Rank	2019 Rank	2020 Rank	Affectionate With Family	...	Coat Length	Openness To Strangers	Playfulness Level	Watchdog/Protective Nature	Adaptability Level	Trainability Level	Energy Level	Barking Level	Mental Stimulation Needs	Average Rank
0	Retrievers (Labrador)	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	5	...	0	5	5	3	5	5	5	3	4	1.000
1	German Shepherd Dogs	2.0	2.0	2.0	2.0	2.0	2.0	2.0	3.0	5	...	1	3	4	5	5	5	5	3	5	2.125
2	Retrievers (Golden)	3.0	3.0	3.0	3.0	3.0	3.0	3.0	4.0	5	...	1	5	4	3	5	5	3	1	4	3.125
3	Beagles	4.0	5.0	5.0	5.0	6.0	6.0	7.0	7.0	3	...	0	3	4	2	4	3	4	4	4	5.625
4	Bulldogs	5.0	4.0	4.0	4.0	5.0	5.0	5.0	5.0	4	...	0	4	4	3	3	4	3	2	3	4.625

5 rows × 26 columns

Modelling

#Splitting 70% train, 30% test
X = df.iloc[:,9:]
X = X.drop('Average Rank', axis=1)
y = np.array(df['Average Rank'])

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=15)

#Create RF Regressor Model with 5000 trees
reg=RandomForestRegressor(n_estimators=5000, random_state=140)

#Fitting
reg.fit(X_train, y_train)

RandomForestRegressor(n_estimators=5000, random_state=140)

#Predict
y_pred = reg.predict(X_test)

#Predicted vs Actual Visualization
px.line(y = [y_pred, y_test], markers=True, title = "Predicted(var0) vs Actual(var1)", labels={"value": "Rank"})

Evaluation

errors = abs(y_pred - y_test)
r2 = r2_score(y_test, y_pred)

print('Average absolute error:', round(np.mean(errors), 2), 'degrees.')
print('R-squared scores:', round(r2,2))
print("MSE  =", round(mean_squared_error(y_test, y_pred),2))
print("RMSE =", round(math.sqrt(mean_squared_error(y_test, y_pred)),2))

Average absolute error: 42.44 degrees.
R-squared scores: 0.21
MSE  = 2342.18
RMSE = 48.4

R-squared score = 21%, meaning the predictors can explain 21% of the variation in the target variable. This result makes the model far below sufficient to be used in real-life situation and should be improved or scrapped entirely.

Improving the Model

#Feature selection
plt.barh(X.columns.values, reg.feature_importances_);

I decide to remove all features below 4% importance.

#Splitting 70% train, 30% test after feature selection
X = df.iloc[:,9:]
X = X.drop(['Mental Stimulation Needs', 'Mental Stimulation Needs', 'Energy Level', 'Average Rank'], axis=1)
y = np.array(df['Average Rank'])

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=15)

#Create RF Regressor model with 5000 trees
reg=RandomForestRegressor(n_estimators=5000, random_state=140)

#Fitting
reg.fit(X_train, y_train)

RandomForestRegressor(n_estimators=5000, random_state=140)

#Predict
y_pred = reg.predict(X_test)

#Predicted vs Actual Visualization
px.line(y = [y_pred, y_test], markers=True, title = "Predicted(var0) vs Actual(var1)", labels={"value": "Rank"})

Evaluation

errors = abs(y_pred - y_test)
r2 = r2_score(y_test, y_pred)

print('Average absolute error:', round(np.mean(errors), 2), 'degrees.')
print('R-squared scores:', round(r2,2))
print("MSE  =", round(mean_squared_error(y_test, y_pred),2))
print("RMSE =", round(math.sqrt(mean_squared_error(y_test, y_pred)),2))

Average absolute error: 41.86 degrees.
R-squared scores: 0.24
MSE  = 2268.18
RMSE = 47.63

R-squared score = 24%, meaning the predictors can explain 24% of the variation in the target variable. While the result has improved, it is still far below the ideal value with at least 70%.